1 / 35

CSci 6971: Image Registration Lecture 3: Images and Transformations January 20, 2004

CSci 6971: Image Registration Lecture 3: Images and Transformations January 20, 2004. Prof. Chuck Stewart, RPI Dr. Luis Ibanez, Kitware. Outline. Images: Types of images Coordinate systems Transformations Similarity transformations in 2d and 3d Affine transformations in 2d and 3d

lotus
Download Presentation

CSci 6971: Image Registration Lecture 3: Images and Transformations January 20, 2004

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. CSci 6971: Image Registration Lecture 3: Images and TransformationsJanuary 20, 2004 Prof. Chuck Stewart, RPI Dr. Luis Ibanez, Kitware

  2. Outline • Images: • Types of images • Coordinate systems • Transformations • Similarity transformations in 2d and 3d • Affine transformations in 2d and 3d • Projective transformations • Non-linear and deformable transformations Lecture 3

  3. Images • Medical (tomographic) images: • CT and MRI • Intensity / video images • Range images Lecture 3

  4. CT: X-Ray Computed Tomography • X-ray projection through body • Reconstruction of interior via computed tomography • A volume of slices • Voxels in each slice are between 0.5 mm and 1.0 mm on a side • Spacing between slices is typically 1.0 and 5.0 mm • Resulting intensities are measured in Hounsfield units • 0 for water • ~ -1000 for air, +1000 for bone Lecture 3

  5. MRI: Magnetic Resonance Images • Magnetic fields: • Nuclei of individual water molecules • Induced fields • Measure the relaxation times of molecules as magnetic fields are changed • Images are recorded as individual slices and as volumes • Slices need not be axial as in CT • MRIs are subject to field distortions • There are many different types of MRIs Lecture 3

  6. Video Images • Pixel intensities depend on • Illumination • Surface orientation • Reflectance • Viewing direction projection • Digitization • By themselves • Pixel dimensions have no physical meaning • Pixel intensities have no physical meaning Lecture 3

  7. Range Images • Measuring depth: • Time of flight • Phase shift • Triangulation • Representation: • (x,y,z) at each pixel • “Point cloud” of (x,y,z) values Brighter pixels are further away Intensity image of same scene Lecture 3

  8. Coordinate Systems • Topics: • Image coordinates • Physical coordinates • Mapping between them • One must be aware of these, especially in the implementation details of a registration algorithm Lecture 3

  9. Image Coordinate Systems • The origin is generally at the corner of the image, often the upper left • Directions: • x axis is across • y axis is down • Differs from matrix coordinates and Cartesian coordinates • Units are integer increments of grid indices (0,0) u v Lecture 3

  10. Physical Coordinates • Origin: • Near center in video images • Defined by scanner for medical images • Directions: • Right-handed coordinate frame • Units are millimeters (medical images) y u x v Lecture 3

  11. Non-Isotropic Voxels in Medical Images • Axial (z) dimension is often greater than within slice (x and y) dimensions • At worst this can be close to an order of magnitude • New scanners are getting close to isotropic dimensions Lecture 3

  12. Scaling from physical to pixel units: parameters are pixels per mm Translation of origin in pixel units Change of Coordinates as Matrix Multiplication • Physical coordinates to pixel coordinates: Lecture 3

  13. Homogeneous Form • Using homogeneous coordinates --- in simplest terms, adding a 1 at the end of each vector --- allows us to write this using a single matrix multiplication: Lecture 3

  14. Transformations • Geometric transformations • Intensity transformations Lecture 3

  15. Transformed moving image, Im’. It is mapped into the coordinate system of If Forward Geometric Transformation “Moving” image, Im “Fixed” image, If Lecture 3

  16. Need to go backward to get pixel value. This will generally not “land” on a single pixel location. Instead you need to do interpolation. Mapping Pixel Values - Going Backwards “Moving” image, Im “Fixed” image, If Lecture 3

  17. Backward Transformation • Because of the above, some techniques estimate the parameters of the “backwards” transformation, T-1: • From the fixed image to the moving image • This can sometimes become a problem if the mapping function is non-invertible • Intensity-based algorithms generally estimate the backwards transformation Lecture 3

  18. Intensity mapping function Intensity mapping parameters Intensity Transformations As Well • When the intensities must be transformed as well, the equations get more complicated: Fortunately, we will not be very concerned with intensity mapping Lecture 3

  19. Transformation Models - A Start • Translation and scaling in 2d and 3d • Similarity transformations in 2d and 3d • Affine transformations in 2d and 3d Lecture 3

  20. Two-component vector Translation and Independent Scaling • We’ve already seen this in 2d: • Transformations between pixel and physical coordinates • Translation and scaling in the retina application • In matrix form: • Where • In homogeneous form: Lecture 3

  21. Similarity Transformations • Components: • Rotation (next slide) • Translation • Independent scaling • “Invariants” • Angles between vectors remain fixed • All lengths scale proportionally, so the ratio of lengths is preserved • Example applications: • Multimodality, intra-subject (same person) brain registration • Registration of range images (no scaling) Lecture 3

  22. Rotations in 2d • Assume (for now) that forward transformation is only a rotation • Image origins are in upper left corner; angles are measured “clockwise” with respect to x axis Im If Lecture 3

  23. Similarity Transformations in 2d • As a result: • Or: Lecture 3

  24. Rotation Matrices and 3d Rotations • The 2d rotation matrix is orthonormal with determinant 1: • In arbitrary dimensions, these two properties hold. • Therefore, we write similarity transformations in arbitrary dimensions as Lecture 3

  25. Representing Rotations in 3d (Teaser) • R is a 3x3 matrix, giving it 9 parameters, but it has only 3 degrees of freedom • Its orthonormality properties remove the other 6. • Question: How to represent R? • Some answers (out of many): • Small angle approximations about each axis • Quaternions Lecture 3

  26. Affine Transformations • Geometry: • Rotation, scaling (each dimension separately), translation and shearing • Invariants: • Parallel lines remain parallel • Ratio of lengths of segments along a line remains fixed. • Example application: • Mosaic construction for images of earth surface (flat surface, relatively non-oblique cameras) Lecture 3

  27. Affine Transformations in 2d • In 2d, multiply x location by arbitrary, but invertible 2x2 matrix: • In general, • Or, in homogeneous form, Lecture 3

  28. Affine Transforms and the SVD • Consider the SVD of the affine matrix A: • (If necessary, go back and review the properties of the SVD.) • Letting • We have • The geometric interpretation of this is striking Lecture 3

  29. Projective Transformations • General linear transformation • Lengths, angle, and parallel lines are NOT preserved • Coincidence, tangency, and complicated ratios of lengths are the invariants • Application: • Mosaics of image taken by camera rotating about its optical center • Mosaic images of planar surface Lecture 3

  30. Algebra of Projective Transformations • Add non-zero bottom row to the homogeneous matrix: • Transformed point location: • Estimation is somewhat more complicated than for affine transformations Lecture 3

  31. Non-Linear Transformations • Transformation using in retinal image mosaics is a non-linear function of the image coordinates: Lecture 3

  32. Deformable Models • Non-global functions • Often PDE-based, e.g.: • Thin-plate splines • Fluid mechanics • Representation • B-splines • Finite elements • Fourier bases • Radial basis functions • Lectures 20-22 will provide an introduction Lecture 3

  33. Summary • Images and image coordinate systems • Watch out for the difference between physical and pixel/voxel coordinates and for non-isotropic voxels! • Transformation models • Forward and backward mapping • Our focus will be almost entirely geometric transformation as opposed to intensity transformations • Transformations: • Similarity • Affine • Projective • Non-linear and deformable Lecture 3

  34. Homework • See handout Lecture 3

  35. Looking Ahead to Lecture 4 • Introduction to intensity-based registration Lecture 3

More Related